OPTIMIZATION OF K-MODE ALGORITHM FOR DATA MINING USING PARTICLE SWARM OPTIMIZATION

Abstract

K-mode is a popular data mining algorithm because of its effective performance in handling categorical data. It has a problem in its methodology in the area of choosing the initial cluster centers for its clustering tasks which usually affects its results. The research proposed a novel PSO K-mode algorithm called PSOKM to improve the performance of K-mode clustering algorithm using PSO. Fitness function was defined based on the structure of K-mode algorithm and weights; the cluster centroids were optimized using PSO. The initial cost for the PSO was taken from K-mode; the weights were picked at random and two centroids from each class were randomly picked. The research used University of California Irvine (UCI) data set and crime data to evaluate the performances of the PSOKM algorithms against conventional K-mode algorithms using metrics such as accuracy, time, sensitivity, specificity and ROC curve. Evaluation result reveals that the PSOKM improved the accuracy of K mode algorithm from 76% to 89.4% using the crime data. The reliability of the algorithms performance was also conducted using UCI data set and the results obtained were compared with the ones from other variant algorithms. The result revealed that the performance of PSOKM were better than that of the respective variants in most cases.